{"paper_id":"2407.14736","title":"Is the difference between deep hedging and delta hedging a statistical arbitrage?","authors":["Pascal François","Geneviève Gauthier","Frédéric Godin","Carlos Octavio Pérez Mendoza"],"paper_date":"2024-07-20","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2407.14736v3","page_url":"https://thequant.space/flowcharts/is-the-difference-between-deep-hedging-and-delta-hedging-a-s/","scores":{"empirical_rigor":8,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper tests whether the difference between deep hedging and delta hedging constitutes a statistical arbitrage in a GARCH‑based market model and shows how the choice of risk measure affects this.","quote":"We test whether such finding remains true in a GARCH-based market model, which is an illustrative case departing from complete market dynamics.","summary":"The recent work of Horikawa and Nakagawa (2024) claims that under a complete market admitting statistical arbitrage, the difference between the hedging position provided by deep hedging and that of the replicating portfolio is a statistical arbitrage. This raises concerns as it entails that deep…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Risk‑free rate r and dividend yield q are constant over the horizon.","type":"market","stated":"explicit","quote":"the annualized continuously compounded risk‑free rate and dividend yield are assumed to be constant with values set to r = 2.66% and q = 1.77%."},{"assumption":"The hedging objective uses Conditional Value‑at‑Risk (CVaRα) as risk measure.","type":"other","stated":"explicit","quote":"we consider the Conditional Value‑at‑Risk (CVaRα) defined as ρ(ξδ T) = E[ξδ T | ξδ T ≥VaRα(ξδ T)], where α ∈(0, 1)"},{"assumption":"The cash amount borrowed in the portfolio remains below a fixed bound B>0.","type":"execution","stated":"explicit","quote":"cash amount borrowed in the portfolio remains below B > 0 (see François et al. (2024))."},{"assumption":"Training data consist of 400,000 independent simulated paths; test data consist of 100,000 independent paths.","type":"data","stated":"explicit","quote":"Agents are trained on training sets of 400,000 independent simulated paths"},{"assumption":"Parameter estimates for the GARCH model are obtained by maximum likelihood on daily S&P 500 data from 2016‑01‑04 to 2020‑12‑31.","type":"data","stated":"explicit","quote":"Parameter estimates are obtained through maximum likelihood on a daily time series of the S&P 500 index extending from January 4, 2016, to December 31, 2020."}]},"datasets":{"status":"available","items":[{"name":"Synthetic simulated paths for training","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"Agents are trained on training sets of 400,000 independent simulated paths with mini‑batch size of 1,000 ..."},{"name":"Synthetic simulated paths for testing","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"Numerical results are obtained from test sets of 100,000 independent paths."},{"name":"S&P 500 daily index series","provider":"S&P 500 index","kind":"market-prices","access":"public","period":"2016-01-04 to 2020-12-31","universe":"S&P 500 index","quote":"Parameter estimates are obtained through maximum likelihood on a daily time series of the S&P 500 index extending from January 4, 2016, to December 31, 2020."}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"When the CVaR confidence level α is low, the deep‑hedging strategy adds a speculative overlay to delta hedging, creating statistical arbitrage opportunities.","kind":"statistical","quote":"we observe that the difference between deep hedging and delta hedging is a speculative overlay if the risk measure considered does not put sufficient relative weight on adverse outcomes."},{"finding":"Using a sufficiently high α for CVaR eliminates the statistical‑arbitrage‑like behavior of the differential position between deep and delta hedging.","kind":"theoretical","quote":"... the difference between deep hedging and delta hedging does not exhibit statistical arbitrage‑like behavior"},{"finding":"The objective function of deep hedging must be carefully selected to avoid speculative behavior and achieve sound hedging performance.","kind":"other","quote":"the objective function of the deep hedging problem must be carefully selected to prevent the hedging agent from abandoning its hedging objective and pursuing speculative behavior"},{"finding":"Risk measures that provide no reward for gains (e.g., semi‑RMSE) can prevent speculation but reduce profitability.","kind":"negative","quote":"A possibility could be to use risk measures that do not provide any reward for gains, such as the semi‑RMSE"}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":17,"citations":{"cited_by_in_corpus":3,"by_independent_authors":2,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":2,"unclassified":1},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":true,"code_url":"https://github.com/cpmendoza/DeepHedging_StatisticalArbitrage","note":null},"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":"2026-11-14T18:05:55Z"},"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/"}