{"paper_id":"2501.03993","title":"Synthetic Data for Portfolios: A Throw of the Dice Will Never Abolish Chance","authors":["Adil Rengim Cetingoz","Charles-Albert Lehalle"],"paper_date":"2025-01-07","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2501.03993v5","page_url":"https://thequant.space/flowcharts/synthetic-data-for-portfolios-a-throw-of-the-dice-will-neve/","scores":{"empirical_rigor":6,"math_complexity":7,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a generative pipeline for multivariate return generation that meets evaluation standards and addresses limitations of generic models.","quote":"In light of these results, we propose a generative pipeline designed to address the limitations discussed earlier. This pipeline begins with appropriate data transformations and decompositions, followed by the use of GANs and parametric models to generate high-dimensional financial multivariate time series.","summary":"Simulation methods have always been instrumental in finance, and data-driven methods with minimal model specification, commonly referred to as generative models, have attracted increasing attention, especially after the success of deep learning in a broad range of fields. However, the adoption of…","strategy":"market generator"},"assumptions":{"status":"available","items":[{"assumption":"The learning accuracy an is inversely related to a power of the initial sample size n.","type":"statistical","stated":"explicit","quote":"Assumption 1. The learning accuracy an is inversely related to a power of the initial sample size n."},{"assumption":"Generating more synthetic data does not improve estimate accuracy if the initial sample bias remains.","type":"statistical","stated":"explicit","quote":"Corollary 2 (Excessive synthetic data generation leads to biased conclusions). If |an| ≥b and U˜n is computed using samples from the learned generative model, lim ˜n→∞P (|U˜n −θ| ≤b) = 0."},{"assumption":"Underlying factor and residual processes are independent with zero mean.","type":"model","stated":"explicit","quote":"The underlying processes are assumed to be independent and have zero mean, i.e. E[F(t)] = 0m and E[Z(t)] = 0d."},{"assumption":"No transaction costs are incurred in the portfolio evaluation.","type":"execution","stated":"explicit","quote":"assume no transaction cost since it is irrelevant for the purpose of evaluation."},{"assumption":"Daily rebalancing of the portfolio is used.","type":"execution","stated":"explicit","quote":"We assume a daily rebalancing and no transaction costs."},{"assumption":"The initial training sample size limits the achievable accuracy of synthetic data estimates.","type":"data","stated":"implied","quote":"It is crucial to recognize that generating more synthetic data does not necessarily improve the accuracy of estimate, on the contrary the more synthetic data are generated, the closer to the initial learning bias the results will be."}]},"datasets":{"status":"available","items":[{"name":"Real US stocks dataset","provider":null,"kind":"market-prices","access":"unknown","period":null,"universe":"US equities","quote":"We evaluate our generative pipeline using a real dataset of US stocks, subjecting it to rigorous tests with a demanding and detail-oriented approach."},{"name":"Synthetic market generator data","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":"US equities","quote":"The market generator gives rather punctual (from an in‑sample point of view) Sharpe ratio estimates"}],"known":[],"sample":{"start":null,"end":null,"market":"US equities","asset_class":"equities","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":false,"detail":null,"quote":"assume no transaction cost since it is irrelevant for the purpose of evaluation.","source":"front_matter","rule":"R3","candidates":{"front_matter":"no","evidence_card":"no","evidence_card_quoted":true},"card_detail":null,"card_quote":"assume no transaction cost since it is irrelevant for the purpose of evaluation.","meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Generating excessive synthetic data leads to biased conclusions and does not improve estimate accuracy.","kind":"theoretical","quote":"Corollary 2 (Excessive synthetic data generation leads to biased conclusions). If |an| ≥b and U˜n is computed using samples from the learned generative model, lim ˜n→∞P (|U˜n −θ| ≤b) = 0."},{"finding":"The market generator’s Sharpe ratio estimates are punctual in‑sample but deviate from out‑of‑sample benchmarks.","kind":"performance","quote":"The market generator gives rather punctual (from an in‑sample point of view) Sharpe ratio estimates, but this is natural since the market..."},{"finding":"120 years of data are required to obtain a tight confidence interval for the Sharpe ratio ground truth.","kind":"statistical","quote":"The right‑hand side of Figure 15 shows that 120 years of data are necessary to get a tight confidence interval on this ground truth."},{"finding":"Long‑only mean‑reversion strategy Sharpe ratios differ between block‑bootstrap and market‑generator samples, indicating model bias.","kind":"performance","quote":"Boxplots of Sharpe ratios of the long‑only mean‑reversion strategy for different look‑back window parameters h backtested on samples from block bootstrap approach (left) and the market generator (right)."}],"reported_performance":{"strategy":"market generator","sharpe":1.09,"sharpe_benchmark":null,"annual_return_pct":19.86,"total_return_pct":null,"max_drawdown_pct":29.48},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"available","supports":[],"contradicts":[],"qualifies":[],"compared_with":22,"citations":{"cited_by_in_corpus":4,"by_independent_authors":4,"how_cited":{"used_as_baseline":0,"uses_method_or_data":1,"extends":0,"discusses_findings":0,"mentions":3,"unclassified":0},"independent_papers":[{"paper":{"paper_id":"2604.07159","title":"SBBTS: A Unified Schrödinger-Bass Framework for Synthetic Financial Time Series","paper_date":"2026-04-08","authors":["Alexandre Alouadi","Grégoire Loeper","Célian Marsala","Othmane Mazhar","Huyên Pham"],"empirical_rigor":6,"page_url":"https://thequant.space/flowcharts/sbbts-a-unified-schr-dinger-bass-framework-for-synthetic-financial-time-series/"},"how":"uses_method_or_data","quote":null}],"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":"2026-11-14T17:47:56Z"},"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/"}