{"paper_id":"2403.00746","title":"A time-stepping deep gradient flow method for option pricing in (rough) diffusion models","authors":["Antonis Papapantoleon","Jasper Rou"],"paper_date":"2024-03-01","paper_type":"theoretical","paper_url":"https://arxiv.org/pdf/2403.00746v2","page_url":"https://thequant.space/flowcharts/a-time-stepping-deep-gradient-flow-method-for-option-pricing/","scores":{"empirical_rigor":6,"math_complexity":8.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a novel deep learning time‑stepping deep gradient flow (TDGF) method for pricing European options in diffusion models.","quote":"We develop a novel deep learning approach for pricing European options in diffusion models, that can efficiently handle high-dimensional problems resulting from Markovian approximations of rough volatility models.","summary":"We develop a novel deep learning approach for pricing European options in diffusion models, that can efficiently handle high-dimensional problems resulting from Markovian approximations of rough volatility models. The option pricing partial differential equation is reformulated as an energy…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The diffusion operator A is symmetric and positive semi‑definite.","type":"model","stated":"explicit","quote":"Assume that A is symmetric and positive semi-definite. Then, Uk solves (3.2) if and only if Uk minimizes …"},{"assumption":"The underlying asset price follows a Markovian diffusion process.","type":"model","stated":"explicit","quote":"Let S denote the price process of a financial asset that evolves according to a (Markovian) diffusion model"},{"assumption":"The time‑stepping scheme uses a backward differentiation formula and a fixed number of time steps N=100.","type":"other","stated":"explicit","quote":"We set the number of time steps equal to N = 100, use 2000 sampling stages in each time step"},{"assumption":"Neural network architecture: 3 layers with 50 neurons per layer, tanh and softplus activations.","type":"other","stated":"explicit","quote":"In the numerical experiments, we have used 3 layers and 50 neurons per layer. The activation functions we have selected are the hyperbolic tangent function, σ1(x) = tanh(x), and the softplus function, σ2(x) = log (ex + 1)"},{"assumption":"Training uses the Adam optimizer with learning rate α = 3 × 10⁻⁴, (β1,β2) = (0.9,0.999) and zero weight decay.","type":"other","stated":"explicit","quote":"In the optimization stage, we use the Adam algorithm [41] with a learning rate α ="}]},"datasets":{"status":"available","items":[{"name":"Heston model parameter set","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"We set the number of time steps equal to N = 100, use 2000 sampling stages in each time step, while in each sampling stage we take 600 samples per dimension"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"synthetic"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"pricing-calibration","findings":[{"finding":"The TDGF method achieves an overall pricing error on the order of 10⁻³, comparable to other deep learning methods but with significantly smaller training times.","kind":"performance","quote":"The empirical results show that the proposed method is accurate, with a small overall error in the order of 10−3 and comparable with other deep learning methods, while the training times are significantly smaller compared to similar methods and do not grow significantly with the dimension of the underlying space."},{"finding":"TDGF training and evaluation times are faster and more consistent than the DGM method.","kind":"performance","quote":"Overall, the TDGF method developed here performs well in the numerical experiments, with an error in the order of 10−3. Moreover, the training and computation (evaluation) times for the TDGF method are faster and more consistent compared to the popular DGM method."},{"finding":"When the dimension of variance processes and correlation increase in the lifted Heston model, the error increases due to the diffusion term becoming smaller.","kind":"negative","quote":"However, when the dimension of the variance processes in the lifted Heston model increases and the correlation increases simultaneously, then the error also increases. This increase in the error is due to the diffusion term becoming smaller, which means that the explicit term in the decomposition (3.1) becomes the domi"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":21,"citations":{"cited_by_in_corpus":3,"by_independent_authors":1,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":1,"unclassified":2},"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":"2026-11-14T18:13:55Z"},"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/"}