{"paper_id":"2509.06702","title":"Nested Optimal Transport Distances","authors":["Ruben Bontorno","Songyan Hou"],"paper_date":"2025-09-08","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2509.06702v1","page_url":"https://thequant.space/flowcharts/nested-optimal-transport-distances/","scores":{"empirical_rigor":7,"math_complexity":8,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes the nested optimal transport (AW) distance as a robust metric for evaluating generative AI on financial time series and introduces a naturally parallelizable algorithm to compute it efficiently.","quote":"We focus on generative AI for financial time series in decision-making applications and employ the nested optimal transport distance, a time-causal variant of optimal transport distance, which is robust to tasks such as hedging, optimal stopping, and reinforcement learning. Moreover, we propose a statistically consistent, naturally parallelizable algorithm for its computation, achieving substantial speedups over","summary":"Simulating realistic financial time series is essential for stress testing, scenario generation, and decision-making under uncertainty. Despite advances in deep generative models, there is no consensus metric for their evaluation. We focus on generative AI for financial time series in…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Adapted empirical measures converge to the underlying distribution as the number of samples N goes to infinity.","type":"statistical","stated":"explicit","quote":"For AW-distances, empirical measures fail to converge, but adapted empirical measures converge to the underlying measure as the number of samples N goes to infinity, which provides statistical guarantee approximating µ with ˆµN in calculating AW-distances; see the proof of Theorem 2 in Appendix A.3. Theorem 2. Let µ ∈P2(RdT ) and ∆N = N −1 dT .","support":"full"},{"assumption":"When the underlying process is Markovian, a modified adapted empirical measure yields faster convergence rates independent of the time horizon T.","type":"model","stated":"explicit","quote":"However, if µ is known to be Markovian, there is a natural Markovian modification of the adapted empirical measure, simply replacing (ˆµN)t+1 q1:t by (ˆµN)t+1 qt , which can be similarly defined as (ˆµN)t+1 q1:t ; see [14]. This modification yields substantially improved statistical convergence rates O(N −1 2d ) that are independent of T; see Theorem 6.1 in [14].","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Ornstein-Uhlenbeck processes","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"Ornstein-Uhlenbeck processes. Let us first consider the Ornstein-Uhlenbeck (OU) processes Xσ = (Xσ t )t≥0, defined by dXσ t = −Xσ t dt + σdWt, X0 = 0, where (Wt)t≥0 is a Brownian motion and σ ∈R+.","support":"full"},{"name":"Fake Brownian Motion","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"Fake Brownian Motion. Let (Wt)t∈[0,1] be a Brownian motion, Z ∼N(0, 1) independent of (Wt)t∈[0,1] and δ ∈(0, 1).","support":"full"}],"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":"unclear","findings":[{"finding":"The proposed parallel algorithm achieves orders-of-magnitude speedups over existing implementations for computing AW distances.","kind":"performance","quote":"Moreover, we propose a statistically consistent, naturally parallelizable algorithm for its computation, achieving substantial speedups over existing approaches. In this paper, we propose a natural parallelization algorithm to calculate AW-distances, which achieves orders-of-magnitude speedups over existing implementations and also behaves statistically consistently.","support":"full"},{"finding":"Numerical values of AW2^2 computed with the Markovian implementation converge to the theoretical values for Ornstein-Uhlenbeck processes.","kind":"statistical","quote":"Denoting by µσ the distribution of ˜Xσ 1:N, we compute AW2 2(µ1, µ3) with randomly generated i.i.d. samples using the Markovian implementation of our algorithm, and evidently, numerical values of AW2 2 converge to its theoretical value; see Figure 1.","support":"full"},{"finding":"The Markovian modification of adapted empirical measures yields substantially improved statistical convergence rates O(N^{-1/2d}) independent of the time horizon.","kind":"theoretical","quote":"This modification yields substantially improved statistical convergence rates O(N −1 2d ) that are independent of T; see Theorem 6.1 in [14].","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":15,"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":2,"extracted_at":"2026-11-14T19:47:55Z"},"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/"}