{"paper_id":"2510.10526","title":"Integrating Large Language Models and Reinforcement Learning for Sentiment-Driven Quantitative Trading","authors":["Wo Long","Wenxin Zeng","Xiaoyu Zhang","Ziyao Zhou"],"paper_date":"2025-10-12","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2510.10526v1","page_url":"https://thequant.space/flowcharts/integrating-large-language-models-and-reinforcement-learning/","scores":{"empirical_rigor":8.5,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a sentiment-driven quantitative trading system using FinGPT for sentiment analysis and integrates these signals with technical indicators via a reinforcement learning algorithm (TD3).","quote":"This research develops a sentiment-driven quantitative trading system that leverages a large language model, FinGPT, for sentiment analysis, and explores a novel method for signal integration using a reinforcement learning algorithm, Twin Delayed Deep Deterministic Policy Gradient (TD3).","summary":"This research develops a sentiment-driven quantitative trading system that leverages a large language model, FinGPT, for sentiment analysis, and explores a novel method for signal integration using a reinforcement learning algorithm, Twin Delayed Deep Deterministic Policy Gradient (TD3). We compare…","strategy":"Rule-based long-short sentiment‑technical combined strategy"},"assumptions":{"status":"available","items":[{"assumption":"Orders are executed at the same day’s closing price (trades filled at the closing auction).","type":"execution","stated":"explicit","quote":"Orders are executed at the same day’s closing price (i.e., trades filled at the closing auction (≈4 p.m. ET). )"},{"assumption":"Zero transaction costs are assumed for the rule‑based strategy by default.","type":"execution","stated":"explicit","quote":"We assume zero transaction costs for the rule-based strategy by default."},{"assumption":"Long‑short strategy starts with an initial investment of zero.","type":"execution","stated":"explicit","quote":"Long-Short Strategy with initial investment of 0."},{"assumption":"News articles released after 4 p.m. are dropped to ensure same‑day trading viability.","type":"data","stated":"explicit","quote":"We also dropped news articles released after 4 p.m. on each trading day to ensure the viability of sentiment signals for same-day trading decisions."}]},"datasets":{"status":"available","items":[{"name":"Thomson Reuters news articles","provider":"Thomson Reuters","kind":"text-news","access":"commercial","period":"2018-2025","universe":"44 S&P 500 stocks with most active news coverage","quote":"The news data are sourced from Thomson Reuters, covering the period from 2018 to 2025, and include the 44 S&P 500 stocks selected based on the most active level of news coverage.","known_id":"refinitiv"},{"name":"CRSP daily stock data","provider":"CRSP","kind":"market-prices","access":"commercial","period":"2018-2025","universe":"44 S&P 500 stocks","quote":"The historical price and volume data for the selected universe are daily-level data sourced from CRSP and Bloomberg, spanning also from 2018 to 2025.","known_id":"crsp"},{"name":"Bloomberg market data (prices, volume, VWAP)","provider":"Bloomberg","kind":"market-prices","access":"commercial","period":"2018-2025","universe":"44 S&P 500 stocks","quote":"The technical data include primarily open and close prices, trading volume (measured as number of shares traded per day) obtained from CRSP, and VWAP sourced from Bloomberg.","known_id":"bloomberg"}],"known":[{"id":"bloomberg","name":"Bloomberg","provider":"Bloomberg L.P.","kind":"market-prices","access":"commercial","quote":"The historical price and volume data for the selected universe are daily-level data sourced from CRSP and Bloomberg, spanning also from 2018 to 2025.","mentions":2,"used_in":1},{"id":"crsp","name":"CRSP","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","quote":"The historical price and volume data for the selected universe are daily-level data sourced from CRSP and Bloomberg, spanning also from 2018 to 2025.","mentions":2,"used_in":1},{"id":"refinitiv","name":"Refinitiv / LSEG (Datastream, Eikon, Tick History)","provider":"LSEG","kind":"market-prices","access":"commercial","quote":"The results from both the rule-based and RL-driven strategies indicate that the sen-timent signal extracted from Thomson Reuters and generated by FinGPT provides some added value over strategies based solely on technical indicators.","mentions":1,"used_in":1}],"sample":{"start":"2018","end":"2025","market":"S&P 500 stocks","asset_class":"equities","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":true,"detail":"A flat transaction cost of 5 basis points per trade is applied in the main backtests; a robustness check also uses 10 bps.","quote":"we also evaluate the strategy’s performance under a transaction cost of 5 basis points per trade.","source":"front_matter","rule":"R3","candidates":{"front_matter":"yes","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"A flat transaction cost of 5 basis points per trade is applied in the main backtests; a robustness check also uses 10 bps.","card_quote":"we also evaluate the strategy’s performance under a transaction cost of 5 basis points per trade.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Applying a 5 bps transaction cost reduces the rule‑based strategy’s annualised return from 13.84% to 3.66% (sentiment weight = 0), a drop of more than 10 % absolute.","kind":"performance","quote":"Comparing the two full sample performance tables, we observe a clear deterioration in annualized return once a 5bps transaction cost is applied. For example, under sentiment weight = 0, the annualized return drops from 13.84% to 3.66%."},{"finding":"Sentiment signals generated by FinGPT add incremental value beyond traditional technical indicators, as indicated by positive performance when combined.","kind":"performance","quote":"The results suggest that sentiment signals generated by FinGPT offer value when combined with traditional technical indicators."},{"finding":"Factor decomposition of the TD3 excess returns shows a high market beta (βmktrf = 0.91) and insignificant idiosyncratic alpha after transaction costs.","kind":"statistical","quote":"... yields R2 = 0.65 and an insignificant intercept (p = 0.93), indicating that the market beta explains most of the variation."},{"finding":"Both rule‑based and RL‑driven strategies exhibit high turnover; the RL strategy’s 52 % annual turnover still delivers superior risk‑adjusted returns despite transaction costs.","kind":"performance","quote":"the RL agent delivers a Sharpe ratio of 1.38 despite a 52 % annual turnover even after imposing a conservatively high-level transaction cost of 10bps."}],"reported_performance":{"strategy":"Rule-based long-short sentiment‑technical combined strategy","sharpe":0.7475,"sharpe_benchmark":null,"annual_return_pct":13.84,"total_return_pct":null,"max_drawdown_pct":37.32},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":31,"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":"2026-11-14T17:24:55Z"},"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/"}