{"paper_id":"2407.21138","title":"Enhancing Deep Hedging of Options with Implied Volatility Surface Feedback Information","authors":["Pascal François","Geneviève Gauthier","Frédéric Godin","Carlos Octavio Pérez Mendoza"],"paper_date":"2024-07-30","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2407.21138v2","page_url":"https://thequant.space/flowcharts/enhancing-deep-hedging-of-options-with-implied-volatility-su/","scores":{"empirical_rigor":7.5,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a dynamic deep‑hedging scheme for S&P 500 options that incorporates implied volatility surface information via a deep policy‑gradient reinforcement‑learning algorithm.","quote":"We present a dynamic hedging scheme for S&P 500 options, where rebalancing decisions are enhanced by integrating information about the implied volatility surface dynamics. The optimal hedging strategy is obtained through a deep policy gradient-type reinforcement learning algorithm.","summary":"We present a dynamic hedging scheme for S&P 500 options, where rebalancing decisions are enhanced by integrating information about the implied volatility surface dynamics. The optimal hedging strategy is obtained through a deep policy gradient-type reinforcement learning algorithm. The favorable…","strategy":"RL-based deep hedging with implied volatility surface"},"assumptions":{"status":"available","items":[{"assumption":"The risk‑free rate and dividend yield are constant and estimated from 2016‑2020 data.","type":"market","stated":"explicit","quote":"continuously compounded risk‑free rate and dividend yield are assumed to be constant and are estimated using data from the period 2016 to 2020, with values set to r = 2.66% and q = 1.77%, respectively."},{"assumption":"Underlying asset price follows an NGARCH(1,1) process with Normal Inverse Gaussian innovations.","type":"model","stated":"explicit","quote":"It integrates an NGARCH(1,1) process with Normal Inverse Gaussian (NIG) innovations to capture volatility clusters and reproduce fat tailed asymmetric returns, while borrowing information from the volatility surface to anchor the evolution of the conditional variance process ht,R."},{"assumption":"Transaction costs are proportional and take values κ ∈ {0%, 0.05%, 0.5%, 1%}.","type":"execution","stated":"explicit","quote":"Various levels of proportional transaction cost are considered, namely κ ∈{0%, 0.05%, 0.5%, 1%}."},{"assumption":"Option data are filtered using conventional OptionMetrics exclusion rules.","type":"data","stated":"explicit","quote":"The sample comes from OptionMetrics database, to which the conventional data exclusion filters are applied."},{"assumption":"Initial underlying price is normalized to S0 = 100 and options are European calls with specified strikes and maturities.","type":"model","stated":"explicit","quote":"The initial underlying asset value is normalized to S0 = 100 for simplicity. The options being hedged are assumed to be European call options (Ψ(ST) = max(ST −K, 0)) with maturities T ∈{21, 63, 126} days"},{"assumption":"The neural network architecture combines LSTM cells with feed‑forward layers to mitigate vanishing gradients.","type":"other","stated":"explicit","quote":"The inclusion of LSTM layers mitigates problems related to vanishing gradients."}]},"datasets":{"status":"available","items":[{"name":"OptionMetrics implied‑volatility surface data","provider":"OptionMetrics","kind":"options","access":"commercial","period":"1996-01-04 to 2020-12-31","universe":"S&P 500 options","quote":"act the daily time series of the IV factor coefficients spanning from January 4, 1996, to December 31, 2020.","known_id":"optionmetrics"},{"name":"Simulated market paths from JIVR model","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"The JIVR model is used in subsequent simulation experiments to generate paths of the variables pertaining to market dynamics (St, {βt,i}"}],"known":[{"id":"optionmetrics","name":"OptionMetrics IvyDB","provider":"OptionMetrics","kind":"options","access":"commercial","quote":"As a byproduct, we 5The sample comes from OptionMetrics database, to which the conventional data exclusion filters are applied.","mentions":3,"used_in":1}],"sample":{"start":"2020-12","end":"2023-10","market":"S&P 500 options","asset_class":"options","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":true,"detail":"Proportional costs κ set to 0%, 0.05%, 0.5% or 1% per trade.","quote":"Various levels of proportional transaction cost are considered, namely κ ∈{0%, 0.05%, 0.5%, 1%}.","source":"front_matter","rule":"R3","candidates":{"front_matter":"yes","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Proportional costs κ set to 0%, 0.05%, 0.5% or 1% per trade.","card_quote":"Various levels of proportional transaction cost are considered, namely κ ∈{0%, 0.05%, 0.5%, 1%}.","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"pricing-calibration","findings":[{"finding":"Reinforcement‑learning hedging agents consistently outperform practitioner delta and Leland‑adjusted delta benchmarks across all transaction‑cost levels.","kind":"performance","quote":"These numerical results reveal that RL agents outperform DH-L for all considered state space configurations, as the relative values of the penalty functions are substantially lower than the reference line value of 1."},{"finding":"Using a reduced state space (excluding portfolio value) improves RL performance and reduces training time by about 190%.","kind":"performance","quote":"where RL techniques are applied in the hedging context without considering the portfolio value as a state variable."},{"finding":"Out‑of‑sample backtests show RL agents achieve lower CVaR95% and SMSE metrics than benchmarks, especially when transaction costs increase.","kind":"performance","quote":"RL agents consistently outperform benchmarks in terms of downside risk (observe CVaR95% and SMSE metrics ...)"},{"finding":"Average profitability (Avg P&L) of all strategies rises with option maturity, but risk (CVaR95%, SMSE) also increases.","kind":"performance","quote":"Our results, illustrated in Figure 4, show that the average profitability of hedging strategies increases with option maturity"},{"finding":"Global importance analysis identifies the underlying return variance (hR), long‑term ATM IV level (β1) and time‑to‑maturity slope (β2) as key drivers of risk‑reduction performance.","kind":"other","quote":"key factors such as the underlying asset return conditional variance process (hR), the long‑term ATM IV level (β1) and the time‑to‑maturity slope (β2)."}],"reported_performance":{"strategy":"RL-based deep hedging with implied volatility surface","sharpe":null,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":25,"citations":{"cited_by_in_corpus":6,"by_independent_authors":2,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":1,"unclassified":5},"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_JIVR","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":"2026-11-14T18:05:11Z"},"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/"}