Leveraging image-based generative adversarial networks for time series generation

Generative models for images have gained significant attention in computer vision and natural language processing, due to their ability to generate realistic samples from complex data distributions. To leverage the advances of image-based generative models for the time series domain, we propose a two-dimensional image representation for time series, called the extended intertemporal return plot (XIRP). Our approach captures the intertemporal time series dynamics in a scale-invariant and invertible way, reducing training time and improving sample quality. We benchmark synthetic XIRPs obtained by an off-the-shelf Wasserstein generative adversarial network with gradient penalty against other image representations and models regarding sample similarity, predictive ability, similarity of moments, and forecast enhancement metrics. Our novel, validated image representation for time series consistently and significantly outperforms the state-of-the-art recurrent neural network and diffusion-based generative model in terms of predictive ability. Synthesis results obtained by alternative image-based representations additionally exceed the benchmarks regarding similarity across moments. Further, we introduce an improved stochastic inversion to substantially improve simulation quality regardless of the representation and provide the prospect of transfer potential in other domains.