{"paper_id":"2402.01441","title":"Learning the Market: Sentiment-Based Ensemble Trading Agents","authors":["Andrew Ye","James Xu","Vidyut Veedgav","Yi Wang","Yifan Yu","Daniel Yan","Ryan Chen","Vipin Chaudhary","Shuai Xu"],"paper_date":"2024-02-02","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2402.01441v2","page_url":"https://thequant.space/flowcharts/learning-the-market-sentiment-based-ensemble-trading-agents/","scores":{"empirical_rigor":4,"math_complexity":6.5,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes integrating sentiment analysis with deep reinforcement learning ensemble algorithms for stock trading, creating a dynamic, sentiment‑based agent switching strategy.","quote":"We propose and study the integration of sentiment analysis and deep reinforcement learning ensemble algorithms for stock trading by evaluating strategies capable of dynamically altering their active agent given the concurrent market environment.","summary":"We propose and study the integration of sentiment analysis and deep reinforcement learning ensemble algorithms for stock trading by evaluating strategies capable of dynamically altering their active agent given the concurrent market environment. In particular, we design a simple-yet-effective…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The trading environment assumes liquidity, non‑negative balance, and transaction costs.","type":"execution","stated":"explicit","quote":"In this environment, an agent will aim to maximize their expected cumulative change in portfolio value following several market constraints and assumptions (liquidity, non-negative balance, and transaction costs)."},{"assumption":"Agents are switched when the sentiment score exceeds a predetermined absolute threshold.","type":"model","stated":"explicit","quote":"If the sentiment score for a given period exceeds a predetermined absolute threshold, then we switch to using the agent which currently holds the highest validation score for the given period."},{"assumption":"Sentiment is captured using daily headlines from leading financial news sources and the AFINN‑en‑165 lexicon.","type":"data","stated":"explicit","quote":"To capture market sentiment, we write a script that gathers daily headlines from leading financial news sources (Wall Street Journal, Bloomberg, etc.) and extracts their sentiments using the AFINN‑en‑165 sentiment lexicon."}]},"datasets":{"status":"available","items":[{"name":"Historical stock price data","provider":null,"kind":"market-prices","access":"public","period":"2010-01-01 to 2016-12-31 (training) and ","universe":null,"quote":"All deep‑learned agents were trained on stock data from 1/1/2010 to 12/31/2016 and then evaluated from 01/01/2017 to 01/01/2019."},{"name":"Wall Street Journal Archive articles","provider":"Wall Street Journal","kind":"text-news","access":"public","period":null,"universe":null,"quote":"For sentiment analysis, we used business articles from the Wall Street Journal Archive, with 15 articles gathered daily to generate an approximation of daily financial sentiment."}],"known":[{"id":"bloomberg","name":"Bloomberg","provider":"Bloomberg L.P.","kind":"market-prices","access":"commercial","quote":"3.2 Capturing Market Sentiment To capture market sentiment, we write a script that gathers daily headlines from leading ﬁnancial news sources (Wall Street Journal, Bloomberg, etc.) and extracts their sentiments using the AFINN-en-165 sentiment lexicon.","mentions":2,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"unclear","headline_net_of_costs":null,"detail":null,"quote":null,"source":"conflict","rule":"R4b","candidates":{"front_matter":"no","evidence_card":"yes","evidence_card_quoted":true},"conflict_kind":"polarity","card_detail":"Transaction costs are included as a market constraint in the environment definition.","card_quote":"In this environment, an agent will aim to maximize their expected cumulative change in portfolio value following several market constraints and assumptions (liquidity, non-negative balance, and transaction costs).","meaning":"the platform classifier and the paper-text extraction disagree, so no value is asserted"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Sentiment‑Ensemble achieves a cumulative return of over 40% over the two‑year test period, an annualized return of 18.36%, and the highest Sharpe (1.32) and Sortino (1.87) ratios among all baselines.","kind":"performance","quote":"Our Sentiment‑Ensemble strategy significantly outperforms both the Dow Jones and the ensemble strategy, resulting in a cumulative return of over 40% over the two‑year period and an annualized return of 18.36%"},{"finding":"Dynamic sentiment‑based agent switching provides the primary performance gain over the conventional fixed‑interval ensemble switching.","kind":"performance","quote":"demonstrate the contribution of each component, and find that the majority of our improvement in trading ability can be attributed to dynamic agent switching as expected."},{"finding":"A single DDPG agent underperforms all ensemble‑based methods and even the Dow Jones benchmark.","kind":"performance","quote":"We also note that an individual deep reinforcement strategy (DDPG) falls behind all methods, including the DJIA."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":29,"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":"2026-11-14T18:15:24Z"},"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/"}