{"paper_id":"2407.15536","title":"Calibrating the Heston model with deep differential networks","authors":["Giovanni Amici","Marco Morandotti","Chen Zhang"],"paper_date":"2024-07-22","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2407.15536v2","page_url":"https://thequant.space/flowcharts/calibrating-the-heston-model-with-deep-differential-networks/","scores":{"empirical_rigor":7,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a gradient‑based deep learning framework called a deep differential network (DDN) to calibrate the Heston option pricing model.","quote":"We propose a gradient-based deep learning framework to calibrate the Heston option pricing model (Heston, 1993).","summary":"We propose a gradient-based deep learning framework to calibrate the Heston option pricing model (Heston, 1993). Our neural network, henceforth deep differential network (DDN), learns both the Heston pricing formula for plain-vanilla options and the partial derivatives with respect to the model…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The risk‑free interest rate r is constant.","type":"model","stated":"explicit","quote":"r is the continuously compounded risk‑free interest rate (assumed to be constant)"},{"assumption":"The Heston model parameters (κ, λ, σ, ρ, v0) are constant over the calibration horizon and not observable in the market.","type":"model","stated":"explicit","quote":"is the vector of model parameters, not observable in the market."},{"assumption":"Partial derivatives of the pricing function are obtained by numerical differentiation of the QuantLib pricer because analytical gradients are unavailable.","type":"data","stated":"explicit","quote":"we recover the partial derivatives by numerical differentiation of the QuantLib pricer, given that the semi-analytical partial derivatives are not available in QuantLib."},{"assumption":"Training and test data are generated synthetically by pricing vanilla call options with QuantLib using the Lewis‑framework quadrature.","type":"data","stated":"explicit","quote":"we generate the dataset used for the neural network training by pricing the call options through the QuantLib engine, which efficiently approximates the integrals of the pricing function by means of exponentially‑fitted Gauss‑Laguerre quadrature in the Lewis framework"},{"assumption":"The network architecture uses Xavier Glorot initialization and a dropout rate of 0.2 to mitigate over‑fitting and gradient issues.","type":"other","stated":"explicit","quote":"we regularize the network by dropping out nodes from each layer, setting a dropout rate of 0.2"}]},"datasets":{"status":"available","items":[{"name":"Synthetic call‑option dataset generated with QuantLib","provider":"QuantLib (Lewis framework)","kind":"options","access":"unknown","period":null,"universe":null,"quote":"we generate the dataset used for the neural network training by pricing the call options through the QuantLib engine"}],"known":[{"id":"yahoo-finance","name":"Yahoo Finance","provider":"Yahoo","kind":"market-prices","access":"public","quote":"S&P500 index. Ticker Number of samples Maturity (days) log (K/S0) MSFT 192 [38, 339] [−0.75, 0.15] COST 178 [42, 272] [−0.72, 0.31] BKNG 204 [37, 247] [−0.76, 0.32] SPX 212 [40, 212] [−0.66, 0.31] 4.3 Calibration setting Once the network is suitably trained, we carry out a calibration test on the quotes of options written on the Microsoft, Costco, BKNG stocks and the S&P500 index, respectively (data downloaded from Y","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"unclear"},"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 DDN significantly outperforms non‑differential feedforward neural networks in calibration accuracy on equity markets.","kind":"pricing-calibration","quote":"Extensive tests on selected equity markets show that the DDN significantly outperforms non-differential feedforward neural networks in terms of calibration accuracy."},{"finding":"The DDN reduces computational time dramatically compared with global optimizers that do not use gradient information.","kind":"performance","quote":"In addition, it dramatically reduces the computational time with respect to global optimizers that do not use gradient information."},{"finding":"Calibration with the DDN achieves roughly the same accuracy as the Nelder‑Mead method but requires only a few seconds instead of minutes.","kind":"pricing-calibration","quote":"the DDN achieves roughly the same accuracy of the Nelder‑Mead calibration method"},{"finding":"Training times of the DDN are comparable to a typical feedforward network, showing methodological parsimony.","kind":"performance","quote":"the DDN training times roughly correspond to the training times of a typical feedforward neural network in which the differentiation layer is not employed"}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":29,"citations":{"cited_by_in_corpus":1,"by_independent_authors":1,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":1,"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":"2026-11-14T18:05:41Z"},"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/"}