{"paper_id":"2407.16103","title":"Reinforcement Learning Pair Trading: A Dynamic Scaling approach","authors":["Hongshen Yang","Avinash Malik"],"paper_date":"2024-07-23","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2407.16103v2","page_url":"https://thequant.space/flowcharts/reinforcement-learning-pair-trading-a-dynamic-scaling-appro/","scores":{"empirical_rigor":7,"math_complexity":6.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a reinforcement‑learning based pair‑trading framework with dynamic investment scaling for cryptocurrency markets.","quote":"We combined reinforcement learning with a statistical arbitrage trading technique, pair trading, which exploits the price difference between statistically correlated assets. ... We","summary":"Cryptocurrency is a cryptography-based digital asset with extremely volatile prices. Around USD 70 billion worth of cryptocurrency is traded daily on exchanges. Trading cryptocurrency is difficult due to the inherent volatility of the crypto market. This study investigates whether Reinforcement…","strategy":"RL-based dynamic scaling pair trading (RL2)"},"assumptions":{"status":"available","items":[{"assumption":"Cryptocurrency prices are extremely volatile.","type":"market","stated":"explicit","quote":"Cryptocurrency is a cryptography-based digital asset with extremely volatile prices."},{"assumption":"Pairs are selected based on high Pearson correlation and Engle–Granger cointegration.","type":"statistical","stated":"explicit","quote":"Pair formation criteria are based on Pearson’s correlation and augment the Engle–Granger two-step cointegration test"},{"assumption":"A flat transaction fee of 0.02% per trade applies.","type":"execution","stated":"explicit","quote":"The transaction cost in the experiment is set to 0.02% commission based on Binance’s fee scheme."},{"assumption":"Adjustable investment amount improves profitability and reduces risk.","type":"model","stated":"explicit","quote":"Having another dimension on the investment side should further enhance profitability and reduce risks."},{"assumption":"Trading intervals of 1 minute provide sufficient data granularity.","type":"data","stated":"explicit","quote":"We performed experiments with the developed reinforcement learner on pairs of BTC‑GBP and BTC‑EUR data separated by 1 min intervals (n = 263,520)."},{"assumption":"RL agents can be trained on historical price data without considering market impact or slippage.","type":"execution","stated":"implied","quote":"The environments are designed to guide the agents by rewarding profitable actions and penalizing costly ones, encouraging effective decision‑making for timing and quantity."}]},"datasets":{"status":"available","items":[{"name":"Binance BTC‑GBP and BTC‑EUR 1‑minute price series","provider":"Binance","kind":"market-prices","access":"public","period":"October 2023 – December 2023","universe":"Bitcoin paired with GBP and EUR","quote":"is from October 2023 to November 2023, and the test is in December 2023, with trading intervals of 1 min (121,500 entries)","known_id":"binance"}],"known":[],"sample":{"start":"2023-12","end":"2023-12","market":"BTC-GBP and BTC-EUR","asset_class":"crypto","frequency":"intraday"},"data_resolution":"intraday-bars"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"Flat 0.02% commission per transaction in both directions, based on Binance fee scheme.","quote":"The transaction cost in the experiment is set to 0.02% commission based on Binance’s fee scheme.","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Flat 0.02% commission per transaction in both directions, based on Binance fee scheme.","card_quote":"The transaction cost in the experiment is set to 0.02% commission based on Binance’s fee scheme.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"RL‑based pair trading achieved annualized profits ranging from 9.94% to 31.53%, outperforming the traditional non‑RL method’s 8.33% annualized profit.","kind":"performance","quote":"The traditional non‑RL pair trading technique achieved an annualized profit of 8.33%, while the proposed RL‑based pair trading technique achieved annualized profits from 9.94% to 31.53%, depending upon the RL learner."},{"finding":"RL1 (A2C) produced fewer drawdowns and a higher win/loss action ratio than the benchmark, though its cumulative profit was not consistently higher.","kind":"performance","quote":"... and has a significantly higher win/loss action ratio due to differences in threshold settings."},{"finding":"RL2 (A2C) delivered the strongest profitability despite a lower win/loss ratio, due to its progressive trading strategy.","kind":"performance","quote":"RL2 displays the strongest profitability, despite a lower win/loss action ratio, due to its progressive trading strategy."},{"finding":"Lower transaction costs improve profitability for all techniques, with RL‑based methods trading more frequently when costs are reduced.","kind":"performance","quote":"We can see that trading techniques generally perform better under lower transaction costs. RL‑based techniques tend to perform more trades when the transaction costs are lower."}],"reported_performance":{"strategy":"RL-based dynamic scaling pair trading (RL2)","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":3974.65,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"available","supports":[{"paper":{"paper_id":"2606.09003","title":"Proof of Stake economy under centralized exchanges--a mean field model","paper_date":"2026-06-08","authors":["Wenpin Tang"],"empirical_rigor":5,"page_url":"https://thequant.space/flowcharts/proof-of-stake-economy-under-centralized-exchanges-a-mean-field-model/"},"question":"How do transaction costs influence trading activity and profitability in financial markets?","this_paper_finding":"Lower transaction costs improve profitability for all techniques, with RL‑based methods trading more frequently when costs are reduced.","other_paper_finding":"Higher transaction costs (faster‑growing L) suppress trading activity, leading to a slower increase in the staking ratio and a slightly more decentralized staking profile.","other_paper_quote":"Comparing with Figure 2, we observe: • Staking ratio increases over time, however, the increase slows as L grows faster (from 0.9 to 0.8 and 0.66 at T0 = 2). This can be explained by the fact that high transaction costs suppress trading activities, and hence, limit the staking ratio. • The staking profile decentralizes over time, and is (slightly) more decentralized as L grows faster (from 6.7 to 6.5 and 6.4 at T0 =","note":"Both findings agree that transaction costs affect trading: lower costs boost profitability and trading frequency (A), while higher costs curb activity and slow staking growth (B).","confidence":0.65},{"paper":{"paper_id":"2410.06568","title":"Statistical Arbitrage in Rank Space","paper_date":"2024-10-09","authors":["Y. -F. Li","G. Papanicolaou"],"empirical_rigor":8,"page_url":"https://thequant.space/flowcharts/statistical-arbitrage-in-rank-space/"},"question":"How do transaction‑cost levels affect the profitability of the RL‑based pair‑trading approach and the rank‑space statistical arbitrage strat","this_paper_finding":"Lower transaction costs improve profitability for all techniques, with RL‑based methods trading more frequently when costs are reduced.","other_paper_finding":"The strategy’s profitability is highly sensitive to transaction‑cost levels, stopping profit generation at 5 bps.","other_paper_quote":"The strategy stops to profit with 5 basis points transaction costs due to substantial costs associated with realizing rank returns in continuous time limit.","note":"Both find profitability highly cost‑sensitive: lower costs boost RL profits and increase trade frequency, while the rank‑space method loses profit at 5 bps.","confidence":0.566}],"contradicts":[],"qualifies":[],"compared_with":38,"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":"2026-11-14T18:05:32Z"},"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/"}