{"paper_id":"2412.06417","title":"Systematic comparison of deep generative models applied to multivariate financial time series","authors":["Howard Caulfield","James P. Gleeson"],"paper_date":"2024-12-09","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2412.06417v1","page_url":"https://thequant.space/flowcharts/systematic-comparison-of-deep-generative-models-applied-to-m/","scores":{"empirical_rigor":6,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a systematic approach to test and compare multivariate price return generators and demonstrates the value of conditional DGMs in an implied volatility trading task.","quote":"Our contribution is trifold. First, we introduce a systematic approach to test and compare multivariate price return generators.","summary":"Financial time series (FTS) generation models are a core pillar to applications in finance. Risk management and portfolio optimization rely on realistic multivariate price generation models. Accordingly, there is a strong modelling literature dating back to Bachelier’s Theory of Speculation in…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The models assume a fixed number of instruments n.","type":"model","stated":"explicit","quote":"In the models below we assume 𝑛instruments."},{"assumption":"The error term εt is i.i.d. with mean zero and constant variance.","type":"statistical","stated":"explicit","quote":"is assumed to be i.i.d with mean zero and variance σ2."},{"assumption":"Trades are excluded when the straddle strike is more than 50 basis points from the forward price.","type":"execution","stated":"explicit","quote":"We rule out any trades where the straddle strike is more than 50 basis points from the forward price."},{"assumption":"When trading implied volatility, the second nearest expiry is used to avoid event volatility.","type":"execution","stated":"explicit","quote":"If the stock has an expected earnings event within five days we trade the second nearest expiry."}]},"datasets":{"status":"available","items":[{"name":"NGARCH+ synthetic dataset","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"We only report the results for the most complicated synthetic datasets, NGARCH+ and Heston+."},{"name":"Heston+ synthetic dataset","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"We only report the results for the most complicated synthetic datasets, NGARCH+ and Heston+."},{"name":"Option data for implied volatility trading","provider":null,"kind":"options","access":"unknown","period":null,"universe":null,"quote":"dataset from the option data. The strike closest to the expiry forward is found."}],"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":"We report the realized profit per day (PnL) (excluding vega profit and transaction fees) in figures 1, 2 and 3.","source":"front_matter","rule":"R3","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":true},"card_detail":null,"card_quote":"We report the realized profit per day (PnL) (excluding vega profit and transaction fees) in figures 1, 2 and 3.","meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"For the NGARCH+ synthetic dataset, RCGAN is the clear best performer among all models.","kind":"performance","quote":"For the NGARCH+ dataset, RCGAN is the clear best performer, however unsurprisingly the GARCH specified models perform well."},{"finding":"For the Heston+ synthetic dataset, GMMN is the best performing DGM model, with RCGAN a close second.","kind":"performance","quote":"GMMN is the best performing DGM model for Heston+ with RCGAN a close second."},{"finding":"In the empirical implied volatility trading task, the generative HAR model outperforms the baseline HAR model.","kind":"performance","quote":"The difference between the generative based HAR and the baseline is stark, with clear outperformance using the generative data."},{"finding":"Network-based features add no value to the generative HAR models.","kind":"negative","quote":"Notably the network features add no value to the generative HAR models."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":19,"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":"2026-11-14T17:50:23Z"},"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/"}