Abstract—Methods for synthetic data generation are extremely valuable nowadays since they allow researchers and practitioners to develop and test their models without the risk and cost associated with using real data. In this paper, we propose a method for the generation of synthetic financial time series. The method adopts time series regimes clustering to perform generative models training on the data from each cluster separately. Also, we suggest the modification of Quantum Generative Adversarial Networks (QuantGAN) architecture that is able to produce synthetic data with frequency characteristics closer to the corresponding real-world time series ones. Our experiments show that (1) synthetic financial time series can be effectively generated by our method; (2) the distribution characteristics of synthetic time series generated by the method are closer to the initial ones in comparison with Fourier Flows and QuantGAN; (3) training the forecasting model on the synthetics generated by the proposed method (Fourier Flows model is used within it) can reduce the forecasting error on the real-world series.
 
Keywords—regime clustering, Generative Adversarial Networks (GAN), normalising flows, time series generation, synthetic time series


Copyright © 2023 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.