{"paper_id":"2507.22035","title":"Quantum generative modeling for financial time series with temporal correlations","authors":["David Dechant","Eliot Schwander","Lucas van Drooge","Charles Moussa","Diego Garlaschelli","Vedran Dunjko","Jordi Tura"],"paper_date":"2025-07-29","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2507.22035v1","page_url":"https://thequant.space/flowcharts/quantum-generative-modeling-for-financial-time-series-with-t/","scores":{"empirical_rigor":4,"math_complexity":8.5,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The paper proposes a Wasserstein quantum generative adversarial network (QGAN) that uses a parameterized quantum circuit as a generator (expectation‑value sampler) and a classical convolutional neural network as a discriminator to generate synthetic financial time series.","quote":"IMPLEMENTATION For this paper, we trained a Wasserstein QGAN for generating time series based on training data originating in the time series of the S&P 500 index. We use a hybrid approach, with a classical neural network as a discrimi-nator and a parameterized quantum circuit (PQC) as a generator.","summary":"Quantum generative adversarial networks (QGANs) have been investigated as a method for generating synthetic data with the goal of augmenting training data sets for neural networks. This is especially relevant for financial time series, since we only ever observe one realization of the process…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The expectation values of single‑qubit Pauli‑X and Pauli‑Z operators are interpreted as the log return of the index at each time step.","type":"model","stated":"explicit","quote":"The discrete time series spans from 20 to 40 points in time, where the expectation value of single-qubit Pauli-X and Pauli-Z operators is interpreted as the log return of the value of the index at each time step.","support":"full"},{"assumption":"Matrix product state (MPS) simulations can efficiently approximate longer time series by exploiting linear structures.","type":"model","stated":"explicit","quote":"While full-state sim-ulations are limited to short time intervals due to their exponential scaling, MPS simulations enable us to model longer time series by exploiting their ability to efficiently replicate linear structures such as financial time series.","support":"full"},{"assumption":"Increasing the bond dimension balances computational cost against simulation accuracy.","type":"model","stated":"explicit","quote":"In the MPS simulations, we vary the bond dimension (also referred to as the tensor train rank) to balance compu-tational costs and simulation accuracy.","support":"full"},{"assumption":"Training data are normalized to zero mean and unit variance before model training.","type":"data","stated":"explicit","quote":"(i) We normalize our training data to have a mean of 0 and a variance of 1: rt,(i) := rt −µr σr , (12) where rt is the original log returns, and µr and σr esti-mate the mean and standard deviation of all log returns.","support":"full"},{"assumption":"An inverse Lambert‑W transform is applied to heavy‑tailed log returns to bring them closer to a Gaussian distribution.","type":"data","stated":"explicit","quote":"(ii) As learning a heavy-tailed distribution can be chal-lenging due to a limited number of samples in the tails, we implement the inverse Lambert-W transform on the normalized log returns.","support":"full"},{"assumption":"The Wasserstein distance computed over the aggregated time‑series distribution is sufficient for training the QGAN.","type":"statistical","stated":"explicit","quote":"This approach leverages the PQC architecture to intrinsically capture temporal correlations in the time series, while the training itself is performed solely on the Wasserstein distance computed over the ag-gregated time series distribution.","support":"full"}]},"datasets":{"status":"available","items":[{"name":"S&P 500 index log returns","provider":null,"kind":"market-prices","access":"public","period":null,"universe":"S&P 500","quote":"In this work, we develop a quantum GAN for syn-thetic data generation and examine its ability in gen-erating time series replicating the S&P 500 index, in-cluding its distribution and temporal correlations. An example of a financial time series is the S&P 500 in-dex, which includes 500 of most valuable companies that are listed on US stock exchanges [25].","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"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 trained QGAN generates synthetic time series that match the target distribution and exhibit temporal correlations such as volatility clustering.","kind":"performance","quote":"The QGAN that we trained generate synthetic financial time series that not only match the target distribution but also exhibit the desired temporal correlations, with the quality of each property depending on the hyperparameters and simulation method. Our trained QGANs generate time series that match the desired distributions and exhibiting some of the tempo-ral correlations seen in financial time series, such as in","support":"full"},{"finding":"MPS simulations with higher bond dimensions and more layers enable training of longer time series compared to full‑state simulations.","kind":"performance","quote":"The lat-ter approach allowed us to simulate PQCs with a higher number of layers and qubits, which makes it possible to train the generation of longer time series. Simulating the PQC with full-state simulations and MPS simulations yield different results, with circuit depth and the MPS bond dimension further influencing the performance.","support":"full"},{"finding":"Fidelity between full‑state and MPS approximations increases with bond dimension, indicating improved accuracy.","kind":"statistical","quote":"As shown in Figure 4, the fidelity increases with χ, indicating improved approxima-tion accuracy.","support":"full"},{"finding":"Generated series from the MPS‑simulated QGAN shows positive, decreasing absolute autocorrelation (volatility clustering) closer to the S&P 500 behavior.","kind":"performance","quote":"In contrast to the time series generated with the full-state simulation shown in Figure 5, the absolute autocorrelation (Subfigure (d)) that indicates volatility clustering is positive and decreasing for all time lags.","support":"partial"},{"finding":"Circuit depth and MPS bond dimension further influence the quality of generated distribution and temporal‑correlation metrics.","kind":"performance","quote":"Simulating the PQC with full-state simulations and MPS simulations yield different results, with circuit depth and the MPS bond dimension further influencing the performance.","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":30,"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":true,"code_url":"https://github.com/LucasAugustusvd/Quantum-Finance","note":null},"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":"2026-11-14T19:48:52Z"},"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/"}